Claude for business: your company's information connected to Claude with permissions and logs
Claude for business is most useful when the assistant can read the company's own systems, and Goldenroz does that connection work. We build MCP servers and install an on-prem MCP gateway that links ERP, CRM, SQL and Microsoft 365 data to Claude, read-only, limited per user and logged. We also plan the rollout to teams and train engineers on Claude Code. Goldenroz is independent and is not an Anthropic partner or reseller. Projects start at $25,000 at a fixed price per phase after a free one-hour scoping call.
From $25,000, fixed price per phase
Updated 2026-11-14. By Shai Goldenberg
What usually goes wrong
- Seats bought, usage flatLicences were assigned and people use the assistant to tidy emails. It knows nothing about customers, orders or stock, so it cannot help with real work.
- Every team connects its own wayOne team installs a desktop extension, another uploads weekly exports. Nobody holds a list of which data is reachable from the assistant.
- Coding agents without guardrailsEngineers run Claude Code with broad permissions on their laptops. There are no project rules, no blocked commands and no review step.
- Security asked lastThe security lead hears about the pilot after it has started and stops it, because nobody can show what the model can reach.
What is included
- Use and data mapThe teams in scope, the questions they need answered and the systems, tables or endpoints that hold the answers.
- Data boundaryA written list of what Claude may read, what is masked before it is sent and what never leaves the organization. Your security lead approves it.
- MCP connectionOur on-prem gateway for common business systems, or a custom MCP server for an internal system. It installs inside your network with Docker Compose.
- Permissions and audit logEach person sees through Claude only what they may already open in the source system. Every question, query and result is recorded.
- Team rolloutA pilot group, a short written guide per team with worked examples from their own data, and a review of the logs after the first weeks.
- Claude Code for engineeringProject rules, guard hooks, permission limits for unattended runs and a pull request workflow, delivered as Claude Code training.
Claude MCP: how company systems are connected
The Model Context Protocol is the open standard Claude uses to call outside tools. An MCP server publishes a short list of named tools, for example an open-orders lookup, and Claude calls one when a question needs it.
Our gateway applies four rules outside the model. Access is read-only. Queries are written and approved in advance, so the model selects one and fills in values and never writes SQL. Permissions are checked for the person asking on every call. Personal fields are masked before results leave your network.
Source systems stay behind your firewall and their credentials are never sent to the model provider. The build itself is described under MCP and Claude integration.
An independent firm, not a reseller
Buyers searching for Anthropic consulting should know where we stand. Goldenroz is not Anthropic, holds no partner status with it and does not resell Claude. You buy your plan from Anthropic or through a cloud provider, in your own name, and we do not quote plan prices.
Independence has a practical side. The same gateway serves ChatGPT and other MCP clients, so the integration work is not lost if you change or add a model later.
What can be checked about our Claude consulting is the engineering behind it. Shai, our CTO, built and owns the gateway, about 27,000 lines of code with 13 connector types, and runs a WhatsApp ordering bot on Claude in production.
Rolling out Claude to teams, one connection at a time
A rollout starts with one team and one system. Finance asking about receivables, or sales asking about pipeline, is enough to learn which questions people really ask. Each further system is a separate fixed-price phase.
Letting Claude create or change records comes later, as an agent project with approval steps, described on the AI integration services page.
How the work runs
Scoping call
Before the projectWe list the systems, the users and the first questions worth answering. You get a fixed price for phase one.
First connector
Phase 1One system connected read-only in a test environment, with a small catalog of approved queries and real user roles.
Security review
Before productionYour IT or security lead reviews permissions, masking rules and the audit log before any production data is connected.
Production install
Phase 2The gateway runs on your infrastructure. A pilot group uses it from Claude or ChatGPT while we watch the logs.
More systems
Later phasesFurther connectors and queries are added one system at a time, each as a fixed-price phase.
What it costs
Each phase has a fixed price agreed before work starts. The price depends on the number of systems, the number of user roles and whether a custom MCP server is needed.
- AI integration projectFixed price per phaseFrom $25,000
- Larger multi-system platformsScoped per project$100,000 and up
- Scoping callEnds with a written scope and a price$0
- Model subscriptions and usageYour own Claude or ChatGPT accountsPaid to the provider
Who it is not for
- Buying Claude licences. We do not resell them.
- Budgets under $25,000.
- General prompt-writing workshops with no connection to company systems.
Who does the work
Shai leads the technology side: business software, SaaS and mobile apps, payments and escrow systems, Postgres security, AI agents built with Claude Code and MCP, and IT and infrastructure projects with senior infrastructure engineers.
Questions buyers ask
How is the first connection set up?
Choose one team and one system, define what may be read, and connect it through an MCP server with per-user permissions and an audit log. After a pilot, add systems one at a time. We do this as fixed-price phases.
Is Goldenroz an Anthropic partner for Claude for business?
No. Goldenroz is an independent engineering firm. You hold your own Claude account with Anthropic or a cloud provider and pay for it directly. Our work is the connection, the controls and the rollout.
Does our data leave the network?
The databases and their credentials stay inside. Only the results of approved queries, after masking, are sent to the model as context for an answer. The boundary document lists the data that is never sent at all.
What does MCP consulting cost?
Projects start at $25,000 at a fixed price per phase, agreed after a free one-hour scoping call. Model subscriptions and usage are paid by you to the provider.
Can Claude update records in our ERP or CRM?
Not in the first phases. We start read-only so that mistakes cannot change data. Write actions are a later project with approval steps and their own scope.
Guides on this topic
- MCP does not replace REST APIs. It wraps them so AI applications can discover and call them. Compare MCP vs API and see when an MCP server is worth building.
- A RAG chatbot answers from your documents by retrieving relevant passages first. Learn how retrieval, permissions, citations and evaluation work in practice.
- MCP is an open standard that connects AI applications to your tools and data. Learn how hosts, clients and MCP servers work, and what to secure first.
AI by industry
- AI integration
- AI for accounting
- AI app and software development
- AI code review and AI coding training
- AI strategy and readiness
- AI for construction
- AI for education
- AI for enterprise
- AI for finance and banking
- Generative AI and LLM development
- AI for healthcare
- AI for insurance
- AI for law firms
- AI for logistics
- AI for manufacturing
- MCP and Claude integration
- AI for real estate
- AI for retail and ecommerce
- AI for small and mid-size business
- AI
Talk to the person who will do the work
One hour, no sales team. You talk to Shai Goldenberg, the CTO, about your systems or your product, and you leave with a clear next step and a price range.